skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: SU-E-J-133: Autosegmentation of Linac CBCT: Improved Accuracy Via Penalized Likelihood Reconstruction

Abstract

Purpose: To improve the quality of kV X-ray cone beam CT (CBCT) for use in radiotherapy delivery assessment and re-planning by using penalized likelihood (PL) iterative reconstruction and auto-segmentation accuracy of the resulting CBCTs as an image quality metric. Methods: Present filtered backprojection (FBP) CBCT reconstructions can be improved upon by PL reconstruction with image formation models and appropriate regularization constraints. We use two constraints: 1) image smoothing via an edge preserving filter, and 2) a constraint minimizing the differences between the reconstruction and a registered prior image. Reconstructions of prostate therapy CBCTs were computed with constraint 1 alone and with both constraints. The prior images were planning CTs(pCT) deformable-registered to the FBP reconstructions. Anatomy segmentations were done using atlas-based auto-segmentation (Elekta ADMIRE). Results: We observed small but consistent improvements in the Dice similarity coefficients of PL reconstructions over the FBP results, and additional small improvements with the added prior image constraint. For a CBCT with anatomy very similar in appearance to the pCT, we observed these changes in the Dice metric: +2.9% (prostate), +8.6% (rectum), −1.9% (bladder). For a second CBCT with a very different rectum configuration, we observed +0.8% (prostate), +8.9% (rectum), −1.2% (bladder). For a third casemore » with significant lateral truncation of the field of view, we observed: +0.8% (prostate), +8.9% (rectum), −1.2% (bladder). Adding the prior image constraint raised Dice measures by about 1%. Conclusion: Efficient and practical adaptive radiotherapy requires accurate deformable registration and accurate anatomy delineation. We show here small and consistent patterns of improved contour accuracy using PL iterative reconstruction compared with FBP reconstruction. However, the modest extent of these results and the pattern of differences across CBCT cases suggest that significant further development will be required to make CBCT useful to adaptive radiotherapy.« less

Authors:
 [1]
  1. Elekta, Inc, Maryland Heights, MO (United States)
Publication Date:
OSTI Identifier:
22494145
Resource Type:
Journal Article
Resource Relation:
Journal Name: Medical Physics; Journal Volume: 42; Journal Issue: 6; Other Information: (c) 2015 American Association of Physicists in Medicine; Country of input: International Atomic Energy Agency (IAEA)
Country of Publication:
United States
Language:
English
Subject:
60 APPLIED LIFE SCIENCES; ACCURACY; ANATOMY; BLADDER; COMPUTERIZED TOMOGRAPHY; IMAGE PROCESSING; IMAGES; ITERATIVE METHODS; LIMITING VALUES; LINEAR ACCELERATORS; PROSTATE; RADIOTHERAPY; RECTUM

Citation Formats

Chen, Y. SU-E-J-133: Autosegmentation of Linac CBCT: Improved Accuracy Via Penalized Likelihood Reconstruction. United States: N. p., 2015. Web. doi:10.1118/1.4924219.
Chen, Y. SU-E-J-133: Autosegmentation of Linac CBCT: Improved Accuracy Via Penalized Likelihood Reconstruction. United States. doi:10.1118/1.4924219.
Chen, Y. Mon . "SU-E-J-133: Autosegmentation of Linac CBCT: Improved Accuracy Via Penalized Likelihood Reconstruction". United States. doi:10.1118/1.4924219.
@article{osti_22494145,
title = {SU-E-J-133: Autosegmentation of Linac CBCT: Improved Accuracy Via Penalized Likelihood Reconstruction},
author = {Chen, Y},
abstractNote = {Purpose: To improve the quality of kV X-ray cone beam CT (CBCT) for use in radiotherapy delivery assessment and re-planning by using penalized likelihood (PL) iterative reconstruction and auto-segmentation accuracy of the resulting CBCTs as an image quality metric. Methods: Present filtered backprojection (FBP) CBCT reconstructions can be improved upon by PL reconstruction with image formation models and appropriate regularization constraints. We use two constraints: 1) image smoothing via an edge preserving filter, and 2) a constraint minimizing the differences between the reconstruction and a registered prior image. Reconstructions of prostate therapy CBCTs were computed with constraint 1 alone and with both constraints. The prior images were planning CTs(pCT) deformable-registered to the FBP reconstructions. Anatomy segmentations were done using atlas-based auto-segmentation (Elekta ADMIRE). Results: We observed small but consistent improvements in the Dice similarity coefficients of PL reconstructions over the FBP results, and additional small improvements with the added prior image constraint. For a CBCT with anatomy very similar in appearance to the pCT, we observed these changes in the Dice metric: +2.9% (prostate), +8.6% (rectum), −1.9% (bladder). For a second CBCT with a very different rectum configuration, we observed +0.8% (prostate), +8.9% (rectum), −1.2% (bladder). For a third case with significant lateral truncation of the field of view, we observed: +0.8% (prostate), +8.9% (rectum), −1.2% (bladder). Adding the prior image constraint raised Dice measures by about 1%. Conclusion: Efficient and practical adaptive radiotherapy requires accurate deformable registration and accurate anatomy delineation. We show here small and consistent patterns of improved contour accuracy using PL iterative reconstruction compared with FBP reconstruction. However, the modest extent of these results and the pattern of differences across CBCT cases suggest that significant further development will be required to make CBCT useful to adaptive radiotherapy.},
doi = {10.1118/1.4924219},
journal = {Medical Physics},
number = 6,
volume = 42,
place = {United States},
year = {Mon Jun 15 00:00:00 EDT 2015},
month = {Mon Jun 15 00:00:00 EDT 2015}
}
  • Purpose: Combining prior day CBCT contours with STAPLE was previously shown to improve automated prostate contouring. These accurate STAPLE contours are now used to guide the planning CT to pre-treatment CBCT deformable registration. Methods: Six IGRT prostate patients with daily kilovoltage CBCT had their original planning CT and 9 CBCTs contoured by the same physician. These physician contours for the planning CT and each prior CBCT are deformed to match the current CBCT anatomy, producing multiple contour sets. These sets are then combined using STAPLE into one optimal set (e.g. for day 3 CBCT, combine contours produced using the planmore » plus day 1 and 2 CBCTs). STAPLE computes a probabilistic estimate of the true contour from this collection of contours by maximizing sensitivity and specificity. The deformation field from planning CT to CBCT registration is then refined by matching its deformed contours to the STAPLE contours. ADMIRE (Elekta Inc.) was used for this. The refinement does not force perfect agreement of the contours, typically Dice’s Coefficient (DC) of > 0.9 is obtained, and the image difference metric remains in the optimization of the deformable registration. Results: The average DC between physician delineated CBCT contours and deformed planning CT contours for the bladder, rectum and prostate was 0.80, 0.79 and 0.75, respectively. The accuracy significantly improved to 0.89, 0.84 and 0.84 (P<0.001 for all) when using the refined deformation field. The average time to run STAPLE with five scans and refine the planning CT deformation was 66 seconds on a Telsa K20c GPU. Conclusion: Accurate contours generated from multiple CBCTs provided guidance for CT to CBCT deformable registration, significantly improving registration accuracy as measured by contour DC. A more accurate deformation field is now available for transferring dose or electron density to the CBCT for adaptive planning. Research grant from Elekta.« less
  • Purpose: To investigate the performance of a new penalized-likelihood PET image reconstruction algorithm using the 1{sub 1}-norm total-variation (TV) sum of the 1st through 4th-order gradients as the penalty. Simulated and brain patient data sets were analyzed. Methods: This work represents an extension of the preconditioned alternating projection algorithm (PAPA) for emission-computed tomography. In this new generalized algorithm (GPAPA), the penalty term is expanded to allow multiple components, in this case the sum of the 1st to 4th order gradients, to reduce artificial piece-wise constant regions (“staircase” artifacts typical for TV) seen in PAPA images penalized with only the 1stmore » order gradient. Simulated data were used to test for “staircase” artifacts and to optimize the penalty hyper-parameter in the root-mean-squared error (RMSE) sense. Patient FDG brain scans were acquired on a GE D690 PET/CT (370 MBq at 1-hour post-injection for 10 minutes) in time-of-flight mode and in all cases were reconstructed using resolution recovery projectors. GPAPA images were compared PAPA and RMSE-optimally filtered OSEM (fully converged) in simulations and to clinical OSEM reconstructions (3 iterations, 32 subsets) with 2.6 mm XYGaussian and standard 3-point axial smoothing post-filters. Results: The results from the simulated data show a significant reduction in the 'staircase' artifact for GPAPA compared to PAPA and lower RMSE (up to 35%) compared to optimally filtered OSEM. A simple power-law relationship between the RMSE-optimal hyper-parameters and the noise equivalent counts (NEC) per voxel is revealed. Qualitatively, the patient images appear much sharper and with less noise than standard clinical images. The convergence rate is similar to OSEM. Conclusions: GPAPA reconstructions using the 1{sub 1}-norm total-variation sum of the 1st through 4th-order gradients as the penalty show great promise for the improvement of image quality over that currently achieved with clinical OSEM reconstructions.« less
  • Purpose: To compare and quantify respiratory motion artifacts in images from free breathing 4D-CT-on-Rails(CTOR) and those from MV-Cone-beam-CT(MVCB) and facilitate respiratory motion guided radiation therapy. Methods: 4D-CTOR: Siemens Somatom CT-on-Rails system with Anzai belt loaded with pressure sensor load cells. 4D scans were performed in helical mode, pitch 0.1, gantry rotation time 0.5s, 1.5mm slice thickness, 120kVp, 400 mAs. Normal and fast breathing (>12rpm) scanning protocols were investigated. Helical scan, AIP(average intensity projection) and MIP(maximum intensity projection) were generated from 4D-CTOR scans with amplitude sorting into 10 phases.MVCB: Siemens Artiste diamond view(1MV)MVCB was performed with 5MU thorax protocol with 60more » second of full rotation.Phantom: Anzai AZ-733V respiratory phantom. The settings were set to normal and resp. modes with repetition rates at 15 rpm and 10 rpm. Surgical clips, acrylic, wooden, rubber and lung density, total six mock-ups were scanned and compared in this study.Signal-to-noise ratio(SNR), contrast-to-noise ratio(CNR) and reconstructed motion volume were compared to different respiratory setups for the mock-ups. Results: Reconstructed motion volume was compared to the real object volume for the six test mock-ups. It shows that free breathing helical in all instances underestimates the object excursions largest to −67.4% and least −6.3%. Under normal breathing settings, MIP can predict very precise motion volume with minimum 0.4% and largest −13.9%. MVCB shows underestimate of the motion volume with −1.11% minimum and −18.0% maximum. With fast breathing, AIP provides bad representation of the object motion; however, the MIP can predict the motion volume with −2.0% to −11.4% underestimate. Conclusion: Respiratory motion guided radiation therapy requires good motion recording. This study shows that regular CTOR helical scans provides bad guidance, 4D CTOR AIP cannot represent the fast breathing pattern, MIP can represent the best motion volume, MVCBCT can only be used for normal breathing with acceptable uncertainties.« less
  • Purpose: CBCT is the current gold standard to verify prone breast patient setup. We investigated in a phantom if non-ionizing localization systems can replace ionizing localization systems for prone breast treatments. Methods: An anthropomorphic phantom was positioned on a prone breast board. Electromagnetic transponders were attached on the left chest surface. The CT images of the phantom were imported to the treatment planning system. The isocenter was set to the center of the transponders. The positions of the isocenter and transponders transferred to the transponder tracking system. The posterior phantom surface was contoured and exported to the optical surface trackingmore » system. A CBCT was taken for the initial setup alignment on the treatment machine. Using the electromagnetic and optical localization systems, the deviation of the phantom setup from the original CT images was measured. This was compared with the difference between the original CT and kV-CBCT images. Results: For the electromagnetic localization system, the phantom position deviated from the original CT in 1.5 mm, 0.0 mm and 0.5 mm in the anterior-posterior (AP), superior-inferior (SI) and left-right (LR) directions. For the optical localization system, the phantom position deviated from the original CT in 2.0 mm, −2.0 mm and 0.1 mm in the AP, SI and LR directions. For the CBCT, the phantom position deviated from the original CT in 4.0 mm, 1.0 mm and −1.0 mm in the AP, SI and LR directions. The measured values from the non-ionizing localization systems differed from those with the CBCT less than 3.0 mm in all directions. Conclusions: This phantom study showed the feasibility of using a combination of non-ionizing localization systems to achieve a similar setup accuracy as CBCT for prone breast patients. This could potentially eliminate imaging dose. As a next step, we are expanding this study to actual patients. This work has been in part supported by Departmental Research Award RODEPT1-JS001, Department of Radiation Oncology, UC Davis Medical Center.« less
  • Purpose: To determine the 6 degree of freedom systematic deviations between 2D/3D and CBCT image registration with various imaging setups and fusion algorithms on the Varian Edge Linac. Methods: An anthropomorphic head phantom with radio opaque targets embedded was scanned with CT slice thicknesses of 0.8, 1, 2, and 3mm. The 6 DOF systematic errors were assessed by comparing 2D/3D (kV/MV with CT) with 3D/3D (CBCT with CT) image registrations with different offset positions, similarity measures, image filters, and CBCT slice thicknesses (1 and 2 mm). The 2D/3D registration accuracy of 51 fractions for 26 cranial SRS patients was alsomore » evaluated by analyzing 2D/3D pre-treatment verification taken after 3D/3D image registrations. Results: The systematic deviations of 2D/3D image registration using kV- kV, MV-kV and MV-MV image pairs were within ±0.3mm and ±0.3° for translations and rotations with 95% confidence interval (CI) for a reference CT with 0.8 mm slice thickness. No significant difference (P>0.05) on target localization was observed between 0.8mm, 1mm, and 2mm CT slice thicknesses with CBCT slice thicknesses of 1mm and 2mm. With 3mm CT slice thickness, both 2D/3D and 3D/3D registrations performed less accurately in longitudinal direction than thinner CT slice thickness (0.60±0.12mm and 0.63±0.07mm off, respectively). Using content filter and using similarity measure of pattern intensity instead of mutual information, improved the 2D/3D registration accuracy significantly (P=0.02 and P=0.01, respectively). For the patient study, means and standard deviations of residual errors were 0.09±0.32mm, −0.22±0.51mm and −0.07±0.32mm in VRT, LNG and LAT directions, respectively, and 0.12°±0.46°, −0.12°±0.39° and 0.06°±0.28° in RTN, PITCH, and ROLL directions, respectively. 95% CI of translational and rotational deviations were comparable to those in phantom study. Conclusion: 2D/3D image registration provided on the Varian Edge radiosurgery, 6 DOF-based system provides accurate target positioning for frameless image-guided cranial stereotactic radiosurgery.« less